app/main.py

import os
from pathlib import Path
from typing import List

from fastapi import FastAPI, HTTPException

from app.config import settings
from app.models import (
    ChatRequest,
    ChatResponse,
    CourseInput,
    CourseProcessError,
    CourseSummary,
    SyllabusProcessResponse,
)
from app.services.gemini_service import GeminiService
from app.services.intent_service import classify_intent
from app.services.pdf_service import chunk_text, fetch_pdf_text
from app.services.student_service import fetch_student_info
from app.vector_store import LocalVectorStore


app = FastAPI(title="GitConnect Chatbot Service", version="0.1.0")
_CONTEXT_LOG_PATH = Path(__file__).resolve().parents[1] / "context.txt"


def _write_context_log(content: str) -> None:
    _CONTEXT_LOG_PATH.write_text(content, encoding="utf-8")


@app.on_event("startup")
def warmup_embedding_model() -> None:
    try:
        GeminiService.preload_embedding_model(settings.embedding_model_name)
    except Exception as exc:
        # Startup should continue even if warmup fails.
        print(f"Embedding warmup skipped due to error: {exc}")


@app.get("/health")
def health() -> dict:
    return {"status": "ok"}


@app.post("/api/syllabus/process", response_model=SyllabusProcessResponse)
def process_syllabus(courses: List[CourseInput]) -> SyllabusProcessResponse:
    if not courses:
        return SyllabusProcessResponse(
            results=[],
            failed=[],
            total_received=0,
            total_processed=0,
            total_failed=0,
        )

    try:
        gemini = GeminiService(
            settings.gemini_api_key,
            settings.gemini_model,
            settings.embedding_model_name,
        )
    except ValueError as exc:
        raise HTTPException(status_code=500, detail=str(exc)) from exc

    vector_store = LocalVectorStore(
        settings.vector_data_dir,
        rag_index_db_url=settings.rag_index_db_url,
        neon_max_retries=settings.neon_max_retries,
        neon_retry_backoff_sec=settings.neon_retry_backoff_sec,
        neon_connect_timeout_sec=settings.neon_connect_timeout_sec,
    )

    results: List[CourseSummary] = []
    failed: List[CourseProcessError] = []

    for course in courses:
        try:
            syllabus_text = fetch_pdf_text(
                str(course.syllabus_url),
                timeout=settings.pdf_timeout_sec,
                max_retries=settings.pdf_max_retries,
                backoff_sec=settings.pdf_retry_backoff_sec,
            )
            if not syllabus_text:
                raise RuntimeError("No text extracted from PDF.")

            chunks = chunk_text(
                syllabus_text,
                chunk_size=settings.rag_chunk_size,
                overlap=settings.rag_chunk_overlap,
            )
            if not chunks:
                raise RuntimeError("Unable to create text chunks from syllabus content.")

            embeddings = [
                gemini.embed_text(chunk, task_type="retrieval_document")
                for chunk in chunks
            ]
            vector_store.upsert_documents(course.semester, course.course_code, chunks, embeddings)

            ai_summary = gemini.summarize_multilingual(course.name, syllabus_text)
            results.append(CourseSummary(course_code=course.course_code, ai_summary=ai_summary))
        except Exception as exc:
            failed.append(CourseProcessError(course_code=course.course_code, error=str(exc)))

    return SyllabusProcessResponse(
        results=results,
        failed=failed,
        total_received=len(courses),
        total_processed=len(results),
        total_failed=len(failed),
    )


@app.post("/api/chat", response_model=ChatResponse)
def chat(req: ChatRequest) -> ChatResponse:
    try:
        gemini = GeminiService(
            settings.gemini_api_key,
            settings.gemini_model,
            settings.embedding_model_name,
        )
    except ValueError as exc:
        raise HTTPException(status_code=500, detail=str(exc)) from exc

    history_text = "\n".join(
        [f"{msg.role}: {msg.content}" for msg in req.history]
    )
    intent, in_scope = classify_intent(req.query)

    if not in_scope:
        reply = (
            "I can help only with education-related queries such as syllabus, attendance, "
            "results, study planning, and course guidance."
        )
        _write_context_log(
            "Intent: out_of_scope\n"
            f"Query: {req.query}\n"
            "LLM Called: no\n"
            f"Response: {reply}\n"
        )
        return ChatResponse(reply_markdown=reply)

    prompt = ""
    chunks_passed = 0

    try:
        if intent in {"attendance", "result"}:
            student_info = fetch_student_info(
                settings.student_performance_url_template,
                req.student_id,
                semester=req.semester,
                intent=intent,
            )
            prompt = f"""
You are a college assistant. Respond in language code: {req.lang_code}.
Return markdown only.

Intent: {intent}
User query: {req.query}

Recent chat history:
{history_text}

Student performance context (authoritative):
{student_info}

Rules:
- Answer only from the provided student performance context.
- If asked for something unavailable in the context, clearly say it is unavailable.
- Be concise and practical.
"""
        elif intent == "syllabus":
            vector_store = LocalVectorStore(
                settings.vector_data_dir,
                rag_index_db_url=settings.rag_index_db_url,
                neon_max_retries=settings.neon_max_retries,
                neon_retry_backoff_sec=settings.neon_retry_backoff_sec,
                neon_connect_timeout_sec=settings.neon_connect_timeout_sec,
            )
            query_embedding = gemini.embed_text(req.query, task_type="retrieval_query")
            hits = vector_store.search(
                req.semester,
                query_embedding,
                top_k=settings.rag_syllabus_top_k,
            )
            hits = hits[: settings.rag_syllabus_top_k]
            chunks_passed = len(hits)
            syllabus_context = "\n\n---\n\n".join(
                [f"[{h.get('course_code', '')}] {h.get('chunk', '')}" for h in hits]
            )

            prompt = f"""
You are a college assistant. Respond in language code: {req.lang_code}.
Return markdown only.

Intent: syllabus
User query: {req.query}

Recent chat history:
{history_text}

Syllabus context (authoritative):
{syllabus_context}

Rules:
- Answer only from the provided syllabus context.
- For unit/module queries, list units clearly with headings/bullets.
- If exact detail is unavailable, state what is missing.
"""
        else:
            prompt = f"""
You are a helpful college assistant. Respond in language code: {req.lang_code}.
Return markdown only.

Intent: other (education-related)
User query: {req.query}

Recent chat history:
{history_text}

Rules:
- Keep the response casual, helpful, and education-focused.
- Do not answer non-education requests.
- If needed, ask a brief clarifying question.
"""

        reply = gemini.generate_markdown(prompt)
    except Exception as exc:
        raise HTTPException(status_code=500, detail=f"LLM response failed: {exc}") from exc

    _write_context_log(
        f"Intent: {intent}\n"
        f"Query: {req.query}\n"
        f"Student ID: {req.student_id}\n"
        f"Semester: {req.semester}\n"
        f"Language: {req.lang_code}\n"
        f"Student endpoint intent param: {intent if intent in {'attendance', 'result'} else 'not_called'}\n"
        f"Student endpoint semester param: {req.semester if intent in {'attendance', 'result'} else 'not_called'}\n"
        f"Syllabus top_k configured: {settings.rag_syllabus_top_k}\n"
        f"RAG chunk size configured: {settings.rag_chunk_size}\n"
        f"RAG chunk overlap configured: {settings.rag_chunk_overlap}\n"
        f"Syllabus chunks passed: {chunks_passed}\n"
        "LLM Called: yes\n"
        "\n--- Prompt Passed To LLM ---\n"
        f"{prompt}\n"
    )

    return ChatResponse(reply_markdown=reply)